Small Scale Pasteurization
Small Scale Pasteurization
The pasteurization process is one of the most common treatments in the food in-
dustries in order to eliminate harmful pathogenic and spoilage microorganism. The pro-
cess consists in maintaining a high temperature stablished during a short period of time
(HTST).
The first task refers to modelling the different elements of the plant. Each one is
modelled using the physical principles resulting a nonlinear model. Different parameters
of the proposed model are later estimated by experimental data. Finally, the complete
model is validated by new experimental data. Linear parameter-varying (LPV) model
represents a class of non-linear systems that can be controlled using powerful linear-like
techniques. As a nonlinear model is proposed to describe the real system, the second task
refers to develop LPV) model.
Results shows that the proposed models fit the experimental data with an error lower
iii
than a 3 %. Advanced controllers could be designed using these models. As a first ap-
proach a PID controller has been designed. The control objective is leading the controlled
variables (temperatures) in order to reach the set point rejecting the disturbances.
Acknowledgements
Thanks to Brais González for offering all his support and knowledge about the pas-
teurization process. His guidance helped me to understand the physical behaviour. Also
I would like to thank Ferran Cortor his help and advises during all the start-up of the
small-scale pasteurization plant.
A special gratitude goes to my family for being such a great example of work, courage
and perseverance, for their support and motivation. They have always raised me up when
I feel falling.
v
Contents
Summary iii
Acknowledgements v
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 System description 7
vii
2.1.2 Heat exchanger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.5 Valves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.6 Pumps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.7 Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3.2 Parameter k2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.3 Parameter k1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 LPV modelling 45
Environmental impact 67
Budget evaluation 69
Concluding remarks 71
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.7 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.8 Further work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Appendix 79
C Vertex matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
D Maple code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
List of Figures
xi
3.2 Experimental relationship between the flow rate F1 and rotor speed of the
feeding pump N1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Experimental relationship between the flow rate F2 and speed of the hot-
water pump N2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.10 Scheme of the different temperatures and flows of the heat exchanger . . . 33
3.12 Evolution of the feeding pump speed (N1 ), the effluent temperature (T2 )
of hot water and heat losses on the heating phase of the exchanger . . . . . 35
3.13 Comparison of the real losses (blue curve) with the losses calculated with
3.18 (red curve) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.17 Evolution of T1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.18 Evolution of T4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Introduction
1.1 Motivation
The pasteurization process is one of the most common treatments in milk, fruit juice
and beverages industries to sterilise and extend their shelf life. It is based on maintaining
a high temperature stablished during a short period of time (HTST). This process is
widely used in food industries and it is considered a complex dynamical system with
nonlinearities. This strategy avoids undesirable quality changes during the process. This
process involves applying heat to the products, during a predetermined period of time
and at certain temperature profile in order to eliminate harmful pathogenic and spoilage
microorganism.
In the past, the control of pasteurization process was based on maintaining the op-
eration stable and reducing the influence of external perturbations. To accomplish these
objectives, a proportional integral derivative (PID) controller was been used with suc-
cessful results [Cai12]. Although the temperature of the pasteurization process can be
controlled with classical control method, changes on the milk temperature are produced
1
2 CHAPTER 1. INTRODUCTION
due to disturbances. To minimize the effect of these phenomena, a higher set point tem-
perature was fixed in order to avoid temperatures below the pasteurization one. The
result of this increase is a loss of energy caused by a greater heating than it is needed.
Further, it can produce alterations in the milks characteristics.
Nowadays, due to the market globalization, the changes in the needs and demands
of the costumers and the increasing interest in the environmental problems, the indus-
tries have to adopt new advanced control techniques in order to remain competitive and
profitable. One of the most used tools is the Model Predictive Control (MPC).
This project deals with the development of a dynamical model for a small-scale
pasteurization plant based on physical principles. As nonlinear equations will be obtained,
an LPV model based on the first ones will be created. The incorporation of a controller
to these models will improve the regulation and track of the temperatures.
To develop a model for a pasteurization process can be very complex considering the
different equations involved and time consuming. Meanwhile, empirical model using real
data makes the development of a model easier for complex process. Since 1998 different
works about pasteurization process can be found in the literature. Ibarrola et al., [JIGS98]
presented a model to describe a high temperature short time pasteurization process. Each
physical element of the plant was identified separately and the complete process was
1.2. STATE OF ART 3
considered as a union of single parts. The model proposed was based on real data and
it incorporated the nonlinearities from varying flow. Along the same sense, Alastruey
et al., [CAGS99] proposed a model for a pasteurization plant based on the structure
of [JIGS98]. The main differences from the previous model were that the fundamental
physical properties of the plant were not neglected, being the blocks of the heat exchanger
of third order. In addition, the delay after the output of the heat exchanger was taken
into account.
Using an empirical model, Wan et al., [WWM12] in 2012 determined the best control
strategy in pasteurization process of a pink guava puree. The control strategy chosen was
the Proportional, Derivative and Integral (PID). In order to improve the control, Khadir
[KRRO] proposed a nonlineal model for the plate of the heat exchanger and the usage
of Model Predictive Control (MPC) to investigate the benefits over the classical control
method (PID).
In 2012, UPC began a new project related to the modelling and control of a pas-
teurization process using a small-scale plant. Caiza [Cai12] developed a black-box model
and designed a predictive control around one working point. Ocampo and Rosich (2015)
followed the investigation and implemented some real-time predictive control schemes
[ROM15]. In this project a new model is developed for the small-scaler pasteurization
plant based on physical principles. Moreover, based on this one, an LPV model is also
created. Both models can be used in a wide operation range.
4 CHAPTER 1. INTRODUCTION
1.3 Objectives
Considering the specific case of the project, the general objective is developed by
achieving the following points:
System description
The plant used in this project is a laboratory version of a real industrial process
manufactured by Armfields Process Plants Trainer: the small-scale pasteurization plant
PTC23 MKII. It represents an industrial process of high-temperature short-time pasteur-
ization (HTST). The goal of this process is to keep the product at the pasteurization
temperature during a minimum period of time stablished.
This system has three main elements: the process unit Figure (2.1), ), the control
console and the software control. The process unit is connected to the control console in
order to monitor on-line the main parameters (temperatures and flows) and control the
system. [Arm10]. The product to be pasteurized is fed either in tank A or B (the valve
SOL2 allows to choose the tank). The peristaltic pump N1 impulses the product from
these tanks to the regeneration phase of the heat exchanger. In this phase, the product
is preheated by the effluent of the holding tube. Then, it passes to the heating phase
in the same heat exchanger where the product achieves the pasteurization temperature.
The product leaves the exchanger at high temperature T4 and flows through the holding
tube to maintain the high temperature (pasteurization temperature) during a certain
time. At the end of the holding tube, there is a valve (SOL1) that opens in case that
7
8 CHAPTER 2. SYSTEM DESCRIPTION
the temperature at this point T1 is higher than the desired. If the valve is open the fluid
returns to the exchanger and passes through two phases. The first one is the regeneration,
where its temperature decreases by preheating the product from tank A and B. The second
phase is cooling, where the goal is to cool the final product using water as a cooling fluid.
On the other hand, when the product temperature is not high enough (T1 is lower than
the desired), the valve SOL1 is closed and it sends the product to the feeding tank again.
The process unit refers to the experimental pilot plant (Figure 2.2). The most import
elements are:
Further, different elements are installed in the pasteurization plant such as valves, pumps
and sensors to control and evaluate the system. These elements are presented in the flow
diagram of Figure 2.1.
The plant PCT23-MKII is provided by two feeding tanks (Figure 2.3) that contain
the product to be heated. They are cylindrical glass tanks with volume of 6 l each one.
The product is fed in these tanks and the flow is controlled by an electronic valve. A
manual valve allows emptying the tank.
10 CHAPTER 2. SYSTEM DESCRIPTION
One of the most important devices of the plant is the heat exchanger. It is a count-
current plate exchanger composed of three different phases as can be seen in Figure 2.4.
Each plate has an area of 8x11.5 cm2 and a thickness of 0.18 cm. In the first phase (central
part Figure 2.4), called regeneration, the product is pre-heated main by the calorific energy
of the end product coming from the holding tube, reducing the energy required in the
next phase. The second phase (right part Figure 2.4), named heating, the product reach
the pasteurization temperature T4 using hot-water as a heating fluid. Finally, the third
phase (left part Figure 2.4) consists of cooling, the end product reduces its temperature
before leaving the process with a cold liquid (in this case water). An image of the heat
exchanger is presented in Figure 2.5.
2.1. PROCESS UNIT 11
The holding tube (Figure 2.6) is a thermally insulated tube with a length of 60 cm
and an internal diameter of 13.2 cm. It allows maintaining the temperature of the product
by a certain time, being this time controlled by the flow. If at the output of the tube the
temperature is still higher than a specific value, the pasteurization will be achieved.
Water is the hot fluid used in the heat exchanger on the heating phase (Figure 2.4).
It is heated in a cylindrical tank (Figure 2.7) with a radio of 7.5 cm and a length of 20.5
cm, by means of an electrical resistor (power of 0-2.5 kW). Depending on the power and
the efficiency of the heat exchanger, the flow can be modified by changing the speed of
the hot-water pump, N2 . The temperature inside the tank T2 is controlled by the power
of the resistor and the speed N2 .
12 CHAPTER 2. SYSTEM DESCRIPTION
2.1.5 Valves
The plant has five solenoid operated valves that can be controlled by the control
console or the PC. They are represented in the flow diagram of Figure 2.1 and can be
described as follow:
2.1.6 Pumps
The process unit also incorporates two peristaltic pumps (Figure 2.8). One impulses
the product from the tanks feed tanks A and B to the heat exchanger (Figure 2.1). The
other one pumps the hot-water from its tank to the heat exchanger (Figure 2.1). Their
speeds belong to the range [0, 400] rpm. These peristatic pumps of the plant are named
as follow:
2.1.7 Sensors
Different sensors are installed in the pilot plant to monitor in certain points the
temperature, level of tanks and flow.
14 CHAPTER 2. SYSTEM DESCRIPTION
Temperature sensor: There are four temperature sensors that monitor the tempera-
ture at four important points (Figure 2.1). These sensors give a measure between 0 and
150◦ C of the following key points:
Tanks A and B are equipped with a level sensor. Tank A level sensor L1 gives a direct
measurement of the product level at tank A in the range [0, 250] mm. In contrast, the
tank B has two fixed float switches (LL and HL) as level sensors. The first one detects
the low level and the other detects the high level.
2.2. CONTROL CONSOLE 15
The control console allows the user to monitor and control the plant either manually
or by the PC. The principal elements of the panel can be seen in Figure 2.9. There are
three groups of potentiometers plus on/off switches that permit to regulate the speed of
the hot-water pump N2 , the feeding pump N1 and the power of the electric resistor P
(1-2-3 in Figure 2.9) manually. The selectors (4-5-6 in Figure 2.9) allow choosing the
operation mode that can be manually, USB or with voltage externally supplied. Further,
the measurement chosen with the selector number 7 can be displayed on the screen.
Finally, the selector number 9 decides between either manually or USB operation of the
solenoids valves (10, 11, 12, 13 and 14 in Figure 2.9). An image of the control console is
16 CHAPTER 2. SYSTEM DESCRIPTION
Figure 2.9: Scheme of the control console implemented in the experimental plant
The software control, allows the communication between the control console and the
PC by using an USB cable. Executing the program ArmSoft PCT23 Process Plant Trainer
(Figure 2.11) the monitoring of the pilot plant in real-time start. The software registers
the data and allows implementing different PID strategies to regulate the temperatures.
2.3. SOFTWARE CONTROL 17
The knowledge of the dynamical response of each one of the elements of the pasteur-
ization plant is fundamental in order to control it. Therefore a fundamental objective of
this Master project is to develop a dynamical model for the PTC23-MKII plant. There
are two ways to develop a model [Lju87]; one is using the physical principles that describe
the system (theoretically) and the other one is using the system identification from exper-
imental data (experimentally). Due to the number of variables of the system and taking
into account that the system can be described by physical process, a parametric model
has been proposed. It is characterized by a finite set of parameters.
To develop the model, the plant has been divided in 5 elements to be modelled: feed-
ing pump, hot-water pump, hot-water tank, holding tube and heat exchanger. Physical
principles based on fundamental laws such as energy balances and heat exchanger design
19
20 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
are used to describe the main processes of the plant [HR04] [JCLA94]. Figure 3.1 shows
a schema of all the elements, its variables and their nomenclature.
In this chapter, the model for each element is presented. Further, the constant
parameters are determined by experimental data and finally the models are validated.1
1
The temperatures, flows, power and speeds presented in the document are time-dependent. To reduce
the complexity of the notation, the time-dependent notation (t) has been omitted
3.1. FEEDING PUMP MODEL 21
The feeding pump (labelled by its speed, N1 ) is the one that impulses the product
to be treated from the feeding tank to the heat exchanger (Figure 3.1). The pump model
correlates the speed of the pump N1 with the fluid flow through the tube F1 . This model
is based on experimental data extracted from a flow meter installed in the plant. Figure
3.2 presents the flow rate F1 for different values of rotor speed N1 , in percentage. The
blue line represents the experimental flow increasing the pump speed in steps of 5 % and
the green line the experimental flow decreasing 5 % the pump speed. Finally, the red line
represents the fitting curve. The graphic shows a linear correlation for speeds higher than
20 % described by the following equation:
F1 = 4.1170N1 − 81.9020,
= kN 1 (N1 − 20),
(3.1)
= 4.1120(N1 − 20),
= 0.0685(N1 − 20).
By 3.1 the value of the gain in static state of the pump can be computed. It can be
considered that the change of the speed immediately produces a change of flow. Further,
the flow (in ml/s) obtained in 3.1 is the same of the mass flow expressed in g/min taking
into account the density (water).
22 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
350
300
250
F1 (ml/min)
200
150
100
F1 increasing
50 F1 decreasing
Fitting curve
0
20 30 40 50 60 70 80 90 100
N1 (%)
Figure 3.2: Experimental relationship between the flow rate F1 and rotor speed of the
feeding pump N1
The hot-water pump impulses the hot-water from the hot-water tank to the heat
exchanger (Figure 3.1). The same procedure as pump N1 has been followed to determine
the correlation between speed of the pump N2 and the flow F2 . However, in this case, a
flow meter is not installed therefore the flow has been experimentally determined at each
speed.
By measuring the volume collected during a period of time. The output tube from
the hot-water tank has been connected to the feeding tank A measuring the volume with
the level sensor. During the experiments the speed N2 has been increased and decreased
in steps of 5 %.
3.2. HOT-WATER PUMP MODEL 23
Figure 3.3 shows the flow rate F2 for different values of rotor speed N2 , in percentage.
The blue line represents the experimental flow increasing the pump speed in steps of 5 %
and the green line the experimental flow decreasing a 5 % the pump speed. Finally, the
red line represents the fitting curve The equation in range 20 % (that is when the water
900
800
700
600
F2 (ml/min)
500
400
300
F2 increasing
200
F2 decreasing
100 Fitting curve
0
20 30 40 50 60 70 80 90 100
N2 (%)
Figure 3.3: Experimental relationship between the flow rate F2 and speed of the hot-water
pump N2
F2 = 9.8223N2 − 142.1700,
= kN 2 (N2 − 20),
(3.2)
= 11.1187(N2 − 20),
= 0.1853(N2 − 20).
As with the feeding pump, it can be considered that the change of the speed immediately
produced the change of flow. The flow obtained in this equation (in ml/s) is the same of
24 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
the mass flow expressed in g/min taking into account the density (water).
The hot-water tank maintains the water at high temperature T2 . This thermal energy
is used in the heat exchanger to the pasteurization process. Hot-water leaves the tank at
a temperature T2 and goes to the heat exchanger. When the exchange is done, it returns
to the tank at a lower temperature T2r . The value of this temperature depends on the
speed of the hot-water pump N2 .
Three different heat transfer processes may be considered in this system as schema-
tized in 3.4:
• The heated system of the tank based on a resistance sunk in the water tank. The
heat transferred (Q0 ) depends on the power resistor applied P . The power is ranged
from 0 W up to 1600 W.
3.3. HOT-WATER TANK MODEL 25
• The loss of heat due to the recirculation flow in the heat exchanger (Q1 ), expressed
by the following equation:
where Ca (J/K) is the specific heat of the water and F2 is the mass flow of hot-
water that flows to the heat exchanger and returns. T2 is the temperature inside
the reactor and T2r is the returned water temperature from the heat exchanger.
• The heat loss to the environment (Q2 ) described by the convection equation [JS07]:
Q2 = U A(T2 − Ta ), (3.4)
where U (Wm2 /K) is the constant of convective heat transfer, A (m2 ) is the area of
the tank and Ta (◦ C) is the room temperature.
Applying an energy balance to the hot-water tank system, the equation can be defined
as:
dTi (t) X
Ci = Eij (t),
dt ij
dT2 (t)
CA = P − F2 Ca (T2 − T2r ) − U A(T2 − Ta ), (3.5)
dt
= P − kN 2 (N2 − 20)Ca (T2 − T2r ) − U A(T2 − Ta ),
The differential equation 3.5 is nonlinear as it has the product of two variables N2
and T2 . The actuators of this element are the power of the electrical resistor P and speed
of the hot-water pump N2 . Further, 3.5 considers three temperatures T2 , T2r (output and
input of the hot-water tank) and Ta (room temperature) that are measurable variables.
Finally, the constant parameters k1 and k2 are determined by adjusting experimental data
and the parameter CA by considering the characteristics and proprieties of the system
26 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
The calorific capacity CA (J/◦ C) is the product of the specific heat (Cp) of the hot-
water and the mass M in (g) that there is in the hot-water tank, i.e,
CA = Cp M. (3.6)
The value of the specific heat is 4.18 J/g◦ C considering pure water at constant temperature
[Wea88]. To calculate the mass of liquid inside the tank , the internal volume of the tank
and the density of the liquid (1000kg/m3 ) are taken into account. Notice that the hot-
water only takes up half of the volume of the tank. Starting from equation 3.6, the new
expression of the calorific capacity takes the following form:
CA = Cp ρV,
(3.7)
= Cp ρπr2 L 2,
where r is the radius in cm of the hot water tank and L is the length also in cm. Substi-
tuting, the final value is 7524 J/◦ C
3.3.2 Parameter k2
To experimentally determine this constant, the recirculation flow has been closed. At
these conditions Q1 is null (Figure 3.4). Then, the equation 3.5 can be rewritten as
dT2 (t)
CA = P − k2 (T2 − Ta ). (3.8)
dt
The power value has been fixed at a 2 % (50 W) and the room has been measured
3.3. HOT-WATER TANK MODEL 27
being 24.4 ◦ C. Taking into account these values, the equation 3.8 can be expressed as:
Z t Z T2
1
dt = 7524 dT2 ,
0 T0 P − k2 (T2 − Ta )
T2
1
t= ln(P − k2 (T2 − Ta )) , (3.9)
−k2 T0
7524 P − k2 (T2 − Ta )
= ln( ,
−k2 P − k2 (T0 − Ta )
To find the value of the parameter k2 a trial and error method has been used to
adjust the experimental data to 3.9. Experimental data T2 vs t is represented in a red
curve in Figure 3.5. Different values of k2 have been tested in order to see which one fits
better to the real data. In Figure 3.5 the different curves are compared. The blue curve
represents k2 = 1.20 J/s◦ C, the green curve k2 = 1.25 J/s◦ C and the light blue curve k2
= 1.30 J/s◦ C. The blue curve is the one that follows better the real data when is reaching
the steady state. This value is considered in the model of the component hot-water tank
3.5.
3.3.3 Parameter k1
Figure 3.6 shows the evolution of experimental temperatures T2 (blue curve) and T2r
(red curve) until they reach the steady state being the values of T2 and T2r at this point
28 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
65
60
55
real data
50 k2 = 1.20
k2 = 1.25
T2 (ºC)
45 k2 = 1.30
40
35
30
25
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
Time (s)
are 68.5 ◦ C and 46.5 ◦ C, respectively and the room temperature 23.5 ◦ C. Considering this
values and equation 3.10, k1 can be calculated and has a final value of 0.4 J/s◦ C. Taking
into account all the parameters calculated in this section, the final equation proposed for
the hot-water tank can be expressed as:
dT2 (t)
7524 = P − 0.4(N2 − 20)(T2 − T2r ) − 1.20(T2 − Ta ). (3.11)
dt
The holding tube is a tube covered with an insulation material. The main finality
of this tube is to maintain the high temperature (pasteurization temperature) during a
period of time. The influent comes from the first stage of the heating exchanger (heating
phase, Figure 3.7) at a flow rate F1 and high temperature T4 . The input temperature
3.4. HOLDING TUBE 29
80
T2
75 T2r
70
65
Temperature (ºC)
60
55
50
45
40
35
0 500 1000 1500 2000 2500 3000 3500
Time (s)
has been considered the same that at the output of the heat exchanger T4 as there is not
a sensor installed at this point. This assumption is based on the proximity of these two
points.
Although the tube is isolated, head loss is expected. Small flow heat (Figure 3.7) is
transferred to the environment producing a decrease of the effluent temperature. Also,
30 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
a delay should be considered depending on the flow rate of the product [GGI01]. This
delay in seconds can be calculated with the following equation:
Volume
τ= ,
F1
(3.12)
82
= .
F1
An experiment has been performed in order to test the precision of 3.12. It consists in
changing suddenly the temperature T4 and determine the effect over T1 . To change the
temperature T4 the speed of the feeding pump N1 has been decreased from 60 % to 40 %.
With these conditions T4 increases because there is less total volume of product to heat.
The theoretical value of the delay τ , obtained with ( 3.12 ) and a product flow of 94
ml/min, is 52 s. Figure 3.8 shows the result of the experiment where the value of τ is 51
s, being the relative error 1.92 %. Assuming a mean temperature Tint inside the holding
45
44.5
44
43.5 T4
Temperature (ºC)
T1
43
42.5
42
41.5
41
40.5
40
1000 1100 1200 1300 1400 1500 1600 1700
Time (s)
dTint1
Q = F1 Cp ∆T + M1 Cp , (3.13)
dt
where ∆ T=T1 (t)-T4 (t-τ ) is the variation of temperature between the output (T1 ) at the
dTint1
instant t and the input (T4 ) at the instant t − τ s and dt
is the variation of the internal
T1 +T4
mean temperature in the tube (Tint1 = 2
). To calculate the mass of liquid M1 , the
internal volume of the tube (82 cm3 ) and the density of the liquid (1000 kg/m3 ) are
considered, i.e. 82 g.
To describe the heat lost in the system Q, the general design equation for head
exchanged is considered [HR12]. Due to not have the characteristics and measures of the
isolation material a generic coefficient of heat transfer U has been considered:
Q = −AU ∆Tml ,
(∆T1 ) − (∆T4 )
= −AU ,
∆T1
ln (3.14)
∆T4
(T1 − Ta ) − (T4 − Ta )
= −AU ,
T1 − Ta
ln
T4 − Ta
where U is the global head transfer coefficient, A is the internal area in contact with the
isolated material that has a value of 0.0248 m2 . Moreover, ∆Tml is the logarithmic mean
temperature. As the mean temperature Tint1 has been considered inside the whole tube,
the equation 3.14 changes into:
Q = −AU ∆T,
The U value has been determined by experimental data. During the experiment
different hot-water pump speeds has been tested in order to evaluate if it influences over
32 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
U . The other actuators N1 and P have been fixed to a constant value 50 % and 175 W,
respectively.
Results have demonstrated that the value of U is positive and remain constant at
different pump speeds (Figure 3.9). Concretely it is 10 W/K m2 . The final equation of
80
60
N1 (%)
40
20
0 50 100 150 200 250 300 350 400 450 500 550
80
60
U (W/m2 ºC)
40
20
0
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
the holding tube taking into account the delay and the equations 3.13 and 3.15 can be
expressed as:
dTint1
F1 Cp (T1 (t) − T4 (t − τ )) + M1 Cp = −U A (Tint1 −Ta ) .(3.16)
dt
The heat exchanger is the device that allows the thermal energy exchange without
mixing the liquids. Heat exchange is controlled by the influent temperature of the fluids
3.5. HEAT EXCHANGER 33
and their mass flow. The heat exchanger of this project had three phases each one with
a different finality: heating, regeneration and cooling (Figure 3.10).
Figure 3.10: Scheme of the different temperatures and flows of the heat exchanger
The first phase, named heating, permitted to head the product from the influent
temperature (Tin ) to the pasteurization temperature. The second phase, regeneration;
recuperates calorific energy from end product to reduce the energy required in the first
phase. Finally, the third phase, called cooling, the temperature of the pasteurized product
is reduced before leaving the system by means of a cold liquid (in this case water).
As mentioned before, the goal of the present project is to control the temperature
T1 , hence, only heating and regeneration phases have been considered. Furthermore, each
phase operates at different conditions therefore, a model of each phase is developed.
34 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
To propose a model of the heating phase, an experimental assay has been performed
focused on this stage. With this objective, the regeneration phase has been stopped
(closing the valve SOL1) therefore, the effluent of the feeding tank is the influent of the
heating phase at the same temperature (Tin = Ta , see Figures 3.1 and 3.11).
On the other side, there is not a temperature sensor in the output of the exchanger.
To monitor T2r (water temperature that returns to the hot-water tank), the sensor of the
product temperature (T3 ) has been moved to that point.
To express the model of the heating phase, an energy balance in dynamic conditions
is proposed:
dTint2
Feeding − product :q = ṁCp ∆Tf eed + M2 Cp ,
dt
Hot − water :q = ṁCp ∆THW T , (3.17)
dTint2
F1 Cp ∆Tf eed + M2 Cp + Qloss1 = F2 Cp ∆THW T .
dt
where ṁ is the mass flow (F1 corresponding to the feed product and F2 to the hot-water)
in g/s, ∆Tf eed = T4 − Tin is the variation of product temperature between output T4
and input Tin being Ti n in this experiment the temperature in the feeding tank Ta . In
dTint2
addition, M2 is the product mass inside the exchanger and is the variation of the
d
∆THW T = T2 − T2r , where T2 is the temperature inside the hot-water tank and T2r the
temperature of the water when it returns to the hot-water tank.
To calculate the product mass M2 , the internal volume of the heating phase (V, with
a value 73.6 cm3 ) and the density of the product (in this case water, 1000 g/l) has been
considered, i.e, 73.6 g.
The heat exchanger is not isolated therefore a heat loss to the environment Qloss1
should be considered. Figure 3.12 shows the influence of the speed of feeding pump (N1 )
and water temperature inside the hot-water tank T2 over the heat loss in the heating
phase of the exchanger. Taking into account the results presented in Figure 3.12, a model
80
N2 (%)
60
40
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
80
T2 (ºC)
60
40
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
600
Losses (W)
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Time (s)
Figure 3.12: Evolution of the feeding pump speed (N1 ), the effluent temperature (T2 ) of
hot water and heat losses on the heating phase of the exchanger
is proposed to determinate the Qloss1 considering the periods of stable temperature. This
36 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
Figure (Figure 3.13) correspond to the validation of the losses equation 3.18. The blue
curve represents the losses obtained from real data and the red curve the losses using 3.18.
Starting from 3.13, the new equation for the heating phase taking into account equation
600
Real data
Model
500
400
300
200
100
0
0 500 1000 1500 2000
Figure 3.13: Comparison of the real losses (blue curve) with the losses calculated with
3.18 (red curve)
dTint2
F1 Cp ∆Tf eed + M2 Cp + (0.1350N1 − 0.4990)(T2 − 27) = F2 Cp ∆THW T . (3.19)
dt
3.5. HEAT EXCHANGER 37
In the regeneration phase, the product from the feeding tank at a temperature Ta is
preheated to reach Tin using as liquid heater the final pasteurized product (Figure 3.14).
With this configuration, part of the energy used in the pasteurization process in reused
in the own system. To extract experimental data of this phase, the heating phase of the
heat exchanger has been stopped. The water of the hot-water tank has been used as the
pasteurized product in order to have a defined value of T1 . This temperature has been
controlled with the power resistor P installed in this tank,
To express the model of the regeneration phase, an energy balance in dynamic con-
ditions is proposed:
dTint3
Feeding − product :q = ṁCp ∆Tf eed + M3 Cp ,
dt
Pasteurized − product :q = ṁCp ∆TReg , (3.20)
dTint3
F1 Cp ∆Tf eed + M3 Cp + Qloss2 = F2 Cp ∆TReg ,
dt
where ∆Tf eed = Tin − Ta is the variation temperature between the output and input
dTint3
product flow. Moreover, M3 is the mass product inside the regeneration phase and dt
is the variation of the internal mean temperature of the product in the regeneration phase
Tint3 = Ta +T
2
in
. Finally, ∆TReg = T1 − T3 is the different temperature of water between
the input and output flow.
38 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
To calculate (M3 ), the internal volume of the regeneration phase (49.68 cm3 ) and the
density of the product are considered, i.e, 49.7 g.
As explained before, the heat exchanger is not isolated therefore a heat loss to the
environment Qloss2 should be contemplated. Figure 3.15 shows that the heat loss do not
seem to be related with the speed of the feeding pump N1 or the temperature T1 and can
be approximated to a constant value of 44 W.
80
N1 (%)
60
40
0 200 400 600 800 1000 1200 1400
80
T1 (ºC)
60
40
0 200 400 600 800 1000 1200 1400
100
Losses (W)
50
0
0 200 400 600 800 1000 1200 1400
Time (s)
The losses are added to the equation 3.20, in the following way:
dTint3
F1 Cp ∆Tf eed + M3 Cp + 44 = F1 Cp ∆TReg . (3.21)
dt
3.6. MODEL VALIDATION 39
In order to validate the resulting models, real data from the pasteurization plant has
been extracted. To evaluate the comparison between the model-obtained temperatures
and the real system temperatures, the Key Performance Indicator (KPI) will be calculated
[KR03] using the expression:
v
u
u1 X N
KP Ii = t (ri (k) − zi (k))2 , (3.22)
N k=1
where N is the number of samples, ri (k) the real temperature of the variable i at the
instant k, zi (k) the temperature of the variable i at the instant k = t calculated by the
continuous model. Finally i is the variable evaluated and k the instant studied.
The software PCT23 allows extracting data in steps of time higher than 1 s. Based
on previous studies with the same small-scale plant, a sampling time of 1 s has been
selected to validate the model [ROM15] [Cai12].
The validation process proposed has been divided in two steps based on changes on
of the different actuators.. The first one step takes the first 250 s and N1 , N2 and P have
been set at 30 %, 30 % and 220 W, respectively. During the second steps the actuators
values have been set at 50 % for N1 , 40 % for N2 and 422 W for P .
The first element validated using integrated approaches is the hot-water tank. With
this methodology the temperature T2 or T2r can be calculated knowing the value of the
other variables. The lectures of the power have not been constants, fluctuating around
the power value stablished, consequently, a filtrate method has been applied. The powers
values resulted after the filtration have been 220 and 422 W.
50
real data
model
45
40
T2 (ºC)
35
30
25
20
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
The second element that has been verified is the holding tube. The temperature T1
(output of the element) has been calculated knowing the value of the actuator N1 and the
variables T4 and Ta . Figure 3.17 shows that the model-obtained temperature (red line)
follows the validation data of T1 (blue line). The KPI is 0.20 ◦ C representing less than a
1 % of error.
As mention in the section 3.5, the heat exchanger has three phases but only two have
been modelled. Figure 3.18 presents the evolution of the output liquid temperature T4
obtained experimentally (blue line) and using the models 3.21 and 3.19 (red line). The
model fits the experimental data at different conditions however it is no able to describe
T4 during an actuators changes as can be seen between 200 and 250s. The KPI obtained
is 0.45 ◦ C representing less than a 2 % of error.
Briefly, Table 3.1 summarizes all KP I indicators from to the dynamic models pro-
3.6. MODEL VALIDATION 41
31
30
29
28
T1 (ºC)
27
26
25
Real data
24 Model
23
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
Analysing the results, the higher value of KPI is lower than 0.5 ◦ C. Taking into
account the range of temperatures that represents a relative error between a 1.5 and 2 %.
These values indicate that the model obtained approaches the data of the real small-scale
pasteurization plant.
42 CHAPTER 3. NONLINEAR MODELLING AND EXPERIMENTAL VALIDATION
Real data
34
Model
32
30
T4 (ºC)
28
26
24
22
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
Finally, Figure 3.19 shows the evolution of the variables T2 , T4 and T1 obtained from
the experimental plant and model-obtained. Also changes on the actuators during the
period studied are depicted.
3.6. MODEL VALIDATION 43
60
N1 (%)
40
20
0 50 100 150 200 250 300 350 400 450 500 550
60
N2 (%)
40
20
0 50 100 150 200 250 300 350 400 450 500 550
600
400
P (W)
200
0
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
50
T2 (ºC)
40
Real data
30
Model
20
0 50 100 150 200 250 300 350 400 450 500 550
35
T4 (ºC)
30
Real data
25
Model
20
0 50 100 150 200 250 300 350 400 450 500 550
30
T1 (ºC)
25 Real data
Model
20
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
LPV modelling
ẋ = A(ψ)x + B(ψ)u,
(4.1)
y = C(ψ)x,
where u is the input, y is the output, x is the vector of state variables, ẋ the derivative
of the state and ψ is an exogenous parameter vector that can be time dependent. It is
assumed that the scheduling parameters can be measure on-line.
Briefly, a LPV model can be defined as a family of linear models where each one
represents the local behavior of a nonlinear plant. The LPV model of a nonlinear plant
can be obtained from the physical equation by means of Jacobian linearization around a
family of working points parametrized. This approach is called linearization scheduling
[CPR12]. In this chapter a family of linear models has been developed and then a LPV
model has been defined.
45
46 CHAPTER 4. LPV MODELLING
The nonlinear models proposed in Chapter 3 have been linearized around several
working points in order to create a family of linear state-space models [Kha02]. Working
points selected should accomplish two main conditions due to the pasteurization process.
The first one is that the temperature T1 (Figure 2.1) should be around the pasteur-
ization temperature and the second one refers to the time that the product maintains this
temperature in the holding tube. With these two conditions the points have been chosen
to cover all the operation range of a HTST (High Temperature Short Time) pasteuriza-
tion process. To determinate the steady values of each variable on the working points,
different constant values of the actuators N1 , N2 and P have been applied until the plant
has achieved the steady state. Moreover, the delay of the holding tube represents the
period of time required for the pasteurization process.
The linearized models are presented by the state-space matrices Ai , Bi and Ci and
their ranges depend on the delay. In order to have always the same matrix range, in this
project the pasteurization time has been considered fixed by feeding pump speed N1 that
means a constant flow F1 .
By 3.12, the value of the feeding pump speed N1 has been calculated considering
the typical pasteurization time ranged between 15-25 s. Finally, a delay or pasteurization
time of 23 s has been proposed with a N1 of 70 %.
Table 4.1 reports the different values of the sensors and actuators at the steady state
of each working point.
In order to have a discrete model, the sampling time Ts has been defined. The
sampling time chosen is 1 s, the same value proposed in previous studies related with the
small-scale pasteurization plant [ROM15] [Cai12].
4.1. FAMILY OF LINEAR LOCAL MODELS 47
In order to verify this sampling time, the model of the element with the lower time
constant T (heat exchanger) has been studied. An empirical rule to select Ts propose
that the maximum value of Ts can be determined as a tenth part of the time constant
T [LMB03]. Figure 4.1 shows the step response for the transfer function of the heat
exchanger. As can be seen in the Figure, the static gain K is 0.7825 ◦ C therefore the time
constant T (time required to achieve a value of 0.632K) is 10.7393 s. and settling time
ts (time required to achieve a value of 0.95K) of 30 s and. Applying the empirical rule
described before, the sampling time should be lower than 1.07 s. Consequently, 1 s can
be employed. In order to develop the state-space model of the real plant, two parts have
been distinguished. The first one includes the hot-water tank and the heat exchanger and
teh second one the holding tube.
The Jacobian linearization based on Taylor series has been applied to the equations
presented in chapter 3 in order to develop a family of local linear models that approximates
the data of the process around the linearization point [LRV05].
The return temperature of the hot-water T2r and the temperature Ti n are not motor-
ized in the experimental plant. Their values are determined by the following expressions:
48 CHAPTER 4. LPV MODELLING
Step Response
0.8
0.7
0.6
0.5
Amplitude
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70
Time (seconds)
T2r = T4 − a.
(4.2)
Tin = T3 − b.
where a and b are constants at each working point. Substituting 4.2, the nonlinear equa-
tions change into:
dT2
F1 = ,
dt (4.3)
P 0.4 1.20
= − (N2 − 20)(T2 − T4 + a) − (T2 − Ta ),
7524 7524 7524
4.1. FAMILY OF LINEAR LOCAL MODELS 49
dT3
F2 = ,
dt
1 4.1120(N1 − 20)Cp (T3 − b − Ta ) 4.1120(N1 − 20)Cp (T1 − T3 )
= + 44 − ,
M4 Cp 60 60
(4.4)
dT4
F3 = ,
dt
2 11.1187(N2 − 20)(T2 − T4 + a) 4.1120(N1 − 20)Cp (T4 − T3 + b)
= − − (4.5)
M2 Cp 60 60
dTin
(0.135N1 − 0.499)(T2 − 27) − ,
dt
dTin
F4 = ,
dt
1 4.1120(N1 − 20)Cp (T1 − T3 ) 4.1120(N1 − 20)Cp (T3 − b − Ta )
= − 44 − ,
M3 Cp 60 60
(4.6)
Applying the Jacobian linearization to 4.3, 4.4 and 4.5 the continuous linearized
state-space model around the working point i has the following structure:
dT2δ (t)
dt
dT3δ (t) T (t) N (t)
2δ 1δ
dt = AAi T (t) + BAi N (t)
3δ 2δ
dT4δ (t) T4δ (t) Pδ (t)
(4.7)
dt
T2δ (t)
y(t) = CAi T3δ (t)
T4δ (t)
where T2δ , T3δ , T4δ , N1δ , N2δ and Pδ are the incremental values around the operating
points of the states and inputs. Matrices AAi , BAi and CAi are expressed as:
50 CHAPTER 4. LPV MODELLING
CAi = 1 1 1
where OPi = {N1opi , N2opi , Popi , T1opi , T2opi , T3opi , T4opi , T4opi , aopi , bopi } are the values of
actuators and variables at the steady state of the working point i.
The continuous state-state model has been transformed to discrete-time using a sam-
pling time Ts of 1 s. The new matrices are given by:
AAdi = eAAi Ts ,
=AAi Ts + I,
Z Ts (4.8)
AAi τ
BAdi = e dτ BAi) ,
0
= Ts BAi .
4.1. FAMILY OF LINEAR LOCAL MODELS 51
The final values of the matrices are presented in Appendix A. The matrices have been
found by using Maple software (see Appendix C).
The holding tube is the element of the plant that presents a delay. As mentioned
before, the delay has been fixed at 23 s. The resulting linearized model can be expressed
by the following transfer function:
T1δ (s)
GT 1 (s) = ,
T4δ (s)
(4.10)
0.08285e−23s
= .
s + 0.08430
Applying a sampling time of 1 s,the discrete transfer function can be expressed as:
T1δ (z)
GT 1 (z) = ,
T4δ (z)
(4.11)
0.07946z −23
= .
z − 0.91920
By the Matlab function ss the state space of the transfer function has been found, resulting
the following expression:
xB (k + 1) = AB xB (k) + BB uB (k),
(4.12)
T1δ (k) = CB xB (k),
where uB (k) = T4δ (k) is the input, xB ∈ <24 is the state vector and AB ∈ <24x24 , BB ∈
<24 , CB ∈ <1x24 . The matrices AB , BB and CB have always the same value because the
linearization only depends on N1 which has a fixed value (see the final values at Appendix
B).
52 CHAPTER 4. LPV MODELLING
The final model involves the two models developed in series, first, hot-water tank
and heat exchanger model and then the holding tube model. This proposal is based on
the fact that the values of the matrices of the holding tube model are constant and do
not depend on the working point. Figure 4.2 shows the scheme of the final model. The
inputs u(k) are N1δ , N2δ and Pδ and the final output is the temperature T1δ . The output
of the first model is T4δ also the input of the second model, therefore the matrix C of the
first model 4.12 changes into:
CAi = 0 0 1
Matrices of the state equation of hot-water tank and heat exchanger, presented in
section 5.1, depend on varying parameters. Therefore, a first approach of a LPV model
for these elements is developed in this section.
As a first approach the matrix BAi (ψ(k)) has been considered invariant. This is based
on working at fixed speeds around each operating point. The incremental inputs N1δ and
N2δ at this conditions are nulls. Therefore, only the state matrix AAi depends on varying
parameter ψ(k) = N2 (k) that is measurable [DR13].
4.2. LPV MODEL 53
PN
with j=1 πj (ψ(k)) = 1 and π(ψ(k)) ≥ 0 where where N is the number of vertex, πj is
the polytopic coordinate of the vertex j and Aj the state-space matrix at that vertex.
By considering the lower and upper bounds of element in ψ(k) for speed ranges [50
%, 80 %] and taking each permutation, 2 vertex models are formed [DR13]. See vertix
matrix at Appendix C. As having only 2 vertex, the polytopic coordinate for the lower
and upper bound, respectively will be calculated as:
ψmax − ψ(k)
π1 (ψ(k)) = ,
ψmax − ψmin (4.14)
π2 (ψ(k)) = 1 − π1 (ψ(k)).
The final LPV model obtained for the hot-water tank and the heat exchanges can be
expressed as:
N
X
xA (k + 1) = πj (ψ(k))Aj xA (k) + BA uB (k),
j=1 (4.15)
T4δ (k) = CB xA (k),
The model has been simulated in the 4 operating points. The values of polytopic coordi-
nates calculated using 4.2at each working point are presented in Table 4.2:
Figure 4.3 shows the evolution of T2 , T4 and T1 around each operating point. The
blue line represents the real data and the red line the temperatures obtained with the
LPV model. The KPI obtained for each temperature are 1.25 ◦ C, 1.25 ◦ C and 1.06 ◦ C for
T2 , T1 and T4 , respectively.
Table 4.3 shows the comparison of the relative error for the nonlinear model (called
54 CHAPTER 4. LPV MODELLING
operating point π1 π2
OP1 1 0
OP2 0.333 0.666
OP3 0 1
OP 0 1
model 1) and the LPV model in series with the holding tube space-state model (model
2).
Model 1 presents a lower relative errors for all the elements compared with the second
model but in both cases lower than 3 %.
4.2. LPV MODEL 55
80
T2 (ºC)
60
Real data
Model
40
0 100 200 300 400 500 600 700
60
T4 (ºC)
50
40
Real data
30 Model
0 100 200 300 400 500 600 700
60
T1 (ºC)
40
Real data
20
Model
0
0 100 200 300 400 500 600 700
Time (s)
Figure 4.3: Evolution of the T2 , T4 and T1
56 CHAPTER 4. LPV MODELLING
Chapter 5
The pasteurization plant presents some perturbations that affect the overall per-
formance. Further, changes in one variable influences to the other ones. Hence, it is
important to design a controller to reach the goals of the process. Although this work has
been focused on the development of a model for the plant,a control strategy using a PID
controller has been designed and simulated as a first approach [ROM15] [WWM12].
KI
GP ID (z) = KP + + KD (1 − z −1 ), (5.1)
1 − z −1
where KP is the proportional gain, KI the integral gain and KD the derivative gain. In
order to avoid the increase of the noise in the experimental plant, the derivative gain
has been considered null. Taking into account this consideration a control loop for each
57
58 CHAPTER 5. PID CONTROL DESIGN AND SIMULATION
The inputs of the smalls-scale pasteurization plant are the speed of the pumps N1
and N2 and the power P of the electrical resistor while the measured outputs are the flow
F1 and the temperatures T1 , T2 , T3 and T4 . For each output parameter, a control loop is
designed, except for T3 as it has not been controlled. A brief description of each control
loop is exposed in the following points:
• Flow F1 controlled by the speed of the pump N1 : One of the goals of a pasteurization
plant is the time that the product remains at certain temperature. This time can
be fixed controlling the flow rate through the holding tube. As proposed in chapter
4, the present study uses a fixed value of 23 s by setting the speed of the speed of
the feeding pump N1 at 70 %. Notice that is a feed-foward control.
• Temperature in the hot-water tank T2 through the power of the electrical resistor P
(Figure 4.5): The hot-water is the thermal energy source used to heat the product.
To a proper system operation, T2 must always be greater than T1 and T4 and enough
to achieve the final temperature desired. Therefore, a controller for this temperature
should be designed in order to guarantee the energy for the process.
• Temperature of the product T1 through the temperature at the output of the heating
phase of the exchanger T4 (Figure 4.6): The temperature T1 (output temperature of
the holding tube) must be the pasteurization temperature stablished in the process.
59
The controller has been designed for a linearized model that works around one of the
operating points stablished in the Chapter 4. The selected working point is characterized
by N1 of 70 % (that represents a flow of 205 ml/min), N2 of 80 % (higher values could
deteriorate the pumps) and a power P of the electrical resistor of 405 W. This power is
high enough to reach a temperature T1 close to 55 [◦C. The operating temperatures with
these actuators values are presented in Table 4.4:
The PI controller designed should permit to reach the reference temperature proposed
without oscillations. Moreover, the settling time has also to be considered. The faster
the desired temperature is reached the fewer product will not be pasteurized. To design
these PI controllers, MATLAB tool sisotool has been used.
60 CHAPTER 5. PID CONTROL DESIGN AND SIMULATION
The model of the hot-water tank is described with the nonlinear equation 3.11 based
on physical principles. The linearization around the working point has been done with
the Taylor series as in [Kuo96]:
dxδ δf δf
= + xδ + yδ , (5.2)
dt δx P0 δy P0
dT2δ
Considering xδ = T2δ , yδ = Pδ and f = dt
isolated from ( 3.11 ), 4.17 can be written as
dT2δ −k1 k2 1
= (N2 − 20) − T2δ + Pδ . (5.3)
dt CA CA CA
After applying the Laplace transform, the transfer function that relates T2δ (s) and Pδ (s)
5.1. CONTROL OF THE TEMPERATURE T2 61
T2δ (s)
GT 2 (s) = ,
Pδ (s)
1
CA
= , (5.4)
−k1 k2
s− (N2 − 20) −
CA CA
0.0001329
= .
s + 0.002824
The discretization of all dynamics is performed using the sampling time of 1 s found in
Chapter 4. The transfer function of the hot-water tank GT 2 , Matlab function c2d has
been used, i.e.,
0.0001327
GT 2 (z) = . (5.5)
z − 0.9962
The PI gains values have been determined considering that the system reaches the desired
temperature of 66 ◦ C without oscillations in the response and reducing the settling time.
The values of the proportional gain KP 1 and integral gain KI1 that satisfies the control
objective are presented in Table 4.5.
Table 5.2: Values of the PI controller gains for the T2 control loop
Parameter Value
KP 1 382.8120
KI1 2.6644
Figure 4.7 represents the evolution of the temperature T2 using the PI controller to
achieve a desire temperature of 66 ◦ C. Using the PI controller, the system reaches the
desired temperature with a lower settling time (240 s) compared to the open loop (1100
s). Also it has a smooth overshoot that represents percentage of a 5 %.
62 CHAPTER 5. PID CONTROL DESIGN AND SIMULATION
70
60
50
40
T2 (ºC)
30
20
10
0
0 200 400 600 800 1000 1200
Time (s)
The temperature T4 is the temperature of the product at the output of the heating
phase of the heat exchanger. The linearization has been done with the Taylor series as
in the section 5.1. and the input has been considered the speed of the hot-water pump
N2 . Applying the linearization and Laplace transform to 3.19 the transfer function can
be expressed as:
T4δ (s)
GT 4 (s) =
,
N2δ (s)
(5.6)
0.07287
= .
s + 0.09312
Being the sampling time of 1 s and using c2d Matlab function, the discrete transfer
function obtained from 4.21 is:
0.0.06958
GT 4 (z) = . (5.7)
z − 0.91110
5.2. CONTROL OF THE TEMPERATURE T4 63
The design parameters of this control loop are to reach the reference without overshoots
and to have a settling time lower than 10 s. Table 4.6 presents the values of the propor-
tional and integral gains that achieve the design conditions.
Table 5.3: Values of the PI controller gains for the T4 control loop
Parameter Value
KP 2 7.6348
KI2 0.6941
Figure 4.8 presents the simulation with the PI controller implemented. It reaches the
desired temperature of 54 ◦ C with a settling time of 7.25 s (lower than the open loop that
was 10.1 s) and remains constant during all the period.
60
50
40
T4 (ºC)
30
20
10
0
0 10 20 30 40 50 60 70 80 90 100
Time (s)
T1δ (s)
GT 1 (s) = ,
T4δ (s)
(5.8)
0.08285 exp−23s
= .
s + 0.08430
Applying c2d Matlab function and the sampling time of 1 s, the discrete function transfer
can be obtained, i.e.,
0.07946z −23
GT 4 (z) = . (5.9)
z − 0.9192
The main objective of controlling the temperature T1 is to achieve the pasteurization
temperature without oscillations or considerable overshoot. In the Table 4.7 are reported
the gains of the PI controller that accomplish the design characteristics.
Table 5.4: Values of the PI controller gains for the T1 control loop
Parameter Value
KP 2 0.1500
KI2 0.0155
Figure 4.9 shows the evolution of T1 using the PI controller designed. It has a delay
of 23 s but then the desired temperature is reached with a settling time of 150 s. The
evolution presents a smooth overshoot of a 1.9 %.
5.3. CONTROL OF THE TEMPERATURE T1 65
60
50
40
T1 (ºC)
30
20
10
0
0 50 100 150 200 250 300 350 400 450 500
Time (s)
Environmental impact
The term environment is used to indicate groups of people, ecosystems, goods, cul-
tures, socio-economic structures, etc. that constitutes the operating framework, not only
the environment intended as physical space.
The small-scale pasteurization plant does not affect the environment with air pollu-
tion but the areas in witch the contents of this project can be applied is the production
of noise pollution. Although the plant has a filtrate equipment to recycle the water there
are some waste is produced.
Having a large scale production will have more environmental effects. There will be
the waste milk (pure milk raw material mixed with water) in dairy wastewater coming
from start-up and shutdown operations performed in the high-temperature, short time
67
68 CHAPTER 6. ENVIRONMENTAL IMPACT
(HTST) pasteurization process. In a large scale production a power resistor will be not
enough to heat up the hot water used. So if a boiler if used then there will be air pollution
due to the gasoil or coal used.
Budget evaluation
This chapter considers the budget to develop the present project considering the cost
of the equipment and licences employed. Human resources cost associated such as tech-
nicians and directors have not been considered.
1. SOFTWARE
Licence MatLab R2014a 1200
Licence Maple 18 0
Licence Texmarker 4.1 0
TOTAL SOFTWARE 1200
2. HARDWARE
HP Pavilion dv6 Notebook Laptop 1200
Pasteurization plant PCT23 MKII 33000
64 bit Card 4.1 6000
TOTAL HARDWARE 40200
69
70 CHAPTER 7. BUDGET EVALUATION
Chapter 8
Concluding remarks
8.1 Conclusions
In the present project two models for real-time control have been developed for a
small-scale pasteurization plant PCT23 MKII. The experimental data has been acquired
by the program ArmSoft PCT23 Process Plant Trainer for the parameter estimation and
validation.
Concerning the nonlinear model based physical principles created, the following con-
clusions can be drawn:
• model validation presents mean quadratic error KP I lower than 0.5 ◦ C with a
relative error of 2 %;
71
72 CHAPTER 8. CONCLUDING REMARKS
tion, this controller has demonstrated to reach the desired temperatures without
oscillations or high overshoots.
• LPV model for a wide operation temperature range is based on the nonlinear model;
• flow F1 has been fixed to avoid changes on the delay and therefore on the matrices
ranges;
• KPI has been calculated by means of experimental data. Values lower than 1.25 C
have been obtained representing a relative error of 3 %;
• the LPV model has a higher relative error compared with the nonlinear model.
8.2 Contributions
The main contributions that have been made in this project are the following:
• Propose a nonlinear and dynamic model for the small-scale pasteurization plant.
Each element of the plant has been modelled considering the physical process that
take place. The constant parameters have been determined by means of exper-
imental data acquired from the plant. The models have been validated by new
experiments at different operation conditions.
• Develop a LPV model based on the nonlinear model. Considering the complexity
of the nonlinear model developed a new linear model has been proposed. As the
Jacobian linearization only allows working around one operating point a family of
linear state-space models have been created. A final LPV model has been defined
as the state-space matrix AAi is function of varying parameters.
8.3. FURTHER WORK 73
• Design a PID controllers for a linearized model. For the desired performance of the
plant, three different temperatures should be controlled. Therefore, a control loop
for each one has been designed tanking into account the interconnections between
element of the plant.
To continue the research related with this project, different ideas are outlined:
• Improve the nonlinear model. The model developed it able to fit experimental
data at different conditions. However when there are changes on the actuators, the
model is not able to take perfectly into account the two possible values during the
integration time.
• Improve the PID controller. PID controller has been simulated in the linearized
model therefore, a new goal could be implement it to the experimental plant and
evaluate the system performance. Moreover, a redesign of the configuration taking
into account the interrelation between temperature T2 and T4 .Also a PID controller
including the derivative action can be designed and tested.
• Improve the state-space model by considering the changes of the delay. In this
project the delay or pasteurization time has been fixed. As this parameter is one of
the most relevant it should be taken into account in the state-space model. There-
fore, matrix range will continuously change depending on the delay conferring com-
plexity to the model.
• Improve the LPV model by considering a varying pump speeds. The LPV has been
validated at fixed speeds hence, a new validation should be done varying the pump
74 CHAPTER 8. CONCLUDING REMARKS
speed during the validation. In this situation state-space matrices AAi and BAi will
be functions of varying parameters.
• Consider the design of Model Predictive Control (MPC) based on the nonlinear
model. MPC has the capability of handling multi-variable control problems taking
into account actuator limitations and physical constraints [ROM15]. The design of
a MPC controller would be the main goal of the future work.
Bibliography
[HR04] David M. Himmelblau and James B. Riggs. Basic Principles and calculations
in chemical engineering. Pearson Education, Inc., 2004.
75
76 BIBLIOGRAPHY
[JIGS98] J.M. Sandoval J.J. Ibarrola, J.C. Guillen and M. Garcia-Sanz. Modelling of a
high temperature short time pasteurization process. Food Control, 9:267–277,
1998.
[Kha02] Hassan K. Khalil. Nonlinear systems. Prentice Hall, Inc, 3 edition, 2002.
[KR03] Mohamed Tarek Khardir and John Ringwood. Linear and nonlinear model
predictive control design for a milk pasteurization plant. Control and Intelli-
gent Systems, 31, 2003.
[Kuo96] Benjamin Kuo. Sistemas de Control Automco. Prentice Hall, Inc, 7 edition,
1996.
[Lju87] Lennart Ljung. System Identification: Theory for the User. Prentice Hall,
1987.
[LMB03] Santiago Garrido Luis Moreno and Carlos Balaguer. Ingeniere Control, Mod-
elado y control de sistemas dincos. Ariell S.A, 2003.
[MGSI01] J.C. Guillen M. GarcSanz and J.J. Ibarrola. Robust controller design for
uncertain systems with variable time delay. Control Engineering Practice,
9:961–972, 2001.
[MKO00] J Ringwood M.T. Khadir, J. Richalet and B. Oconnor. Modelling and predic-
tive control of milk pasteurisation in a plate heat exchanger. Proc. Foodsim,
2000.
[Oga95] Katsuiko Ogata. Discrete-Time Control Systems. Prentice Hall, Inc, 2 edition,
1995.
[ROM15] A. Rosich and C. Ocampo-Martz. Real-time experimental implementation of
predictive control schemes in a small-scale pasteurization plant. Developments
in Model-Based Optimization and Control, 464:255–273, 2015.
[Wea98] Robert C. Weast. CRC Handbook of chemistry and physics. CRC Press, Inc.,
1998.
[WWM12] N. Abdut Aziz S.B. Mohd Noor W.M.F. Wan Mokhtar, F.S. Taip. Process
control of pink guava puree pasteurization process: Simulation and validation
by experiment. International Journal on Advanced Science, Engineering and
Information Technology, 2:31–34, 2012.
Appendices
The matrices at each operational point for hot-water tank and heat exchanger con-
tinuous state-space are:
Operating point 1:
−0.001754 0 0.001595
AcA1 = −0.91991
0 0
0.092879 0.231010 −0.244185
0 −0.001356 0.000133
BA1 = −0.001923
0 0
−0.067769 0.128459 0
Operating point 2:
−0.002818 0 0.002658
AcA2 = −0.91991
0 0
0.193592 0.231010 −0.344898
79
0 −0.000828 0.000133
BA2 = −0.002824
0 0
−0.072313 0.078405 0
Operating point 3:
−0.003349 0 0.003190
AcA3 = −0.91991
0 0
0.243949 0.231010 −0.395255
0 −0.000769 0.000133
BA3 = −0.003284
0 0
−0.081040 0.072866 0
Operating point 4:
−0.003349 0 0.003190
AcA4 = −0.91991
0 0
0.243949 0.231010 −0.395255
0 −0.000942 0.000133
BA4 = −0.003744
0 0
−0.098564 0.089181 0
The matrices for the discrete holding tube state-space model were:
0.9192 0 · · · 0 0
1 0 ··· 0 0
AB = · · · 0 0
0 1
.. .. . . .. ..
. . .
. .
0 0 ··· 1 0
0.2500
0
BA1 = .
..
0
BA1 = 0 · · · 0 0.3178
C Vertex matrices
0.996651 0
0.003190
A2 =
0 0.080090 0
0.243949 0.231010 0.604745
D Maple code
Parameters (1)
Equations (2)
(3)
(4)
(5)
(6)
Matriz_A (7)
0
(8)
0 (9)
(1)
(2)
Matriz_B (3)
1
(4)
7524
0
0 (5)
0 (6)
(7)